Summary

The main objective of the simplicity model of unsupervised categorization is to predict the relative intuitiveness of different classifications of items, based on their similarity. The model derives from the minimum description length framework, which is an algorithmic formalization of Ockham's razor. It generally prefers classifications which maximize within- category similarity, while minimizing between-category similarity.

Description of the model

The simplicity model of categorization can be viewed as one route to the formalization of Rosch and Mervis's (1975) proposal concerning the nature of ‘basic level’ categories, that is, the categories with which we prefer to classify objects (as opposed to corresponding superordinate or subordinate categories). According to Rosch and Mervis, basic level categories are those that maximize within-category similarity and minimize between-category similarity. Pothos and Chater (2002) examined whether this proposal may be suitable for predicting category preference generally, within a computational framework based on the simplicity principle.

The application of the simplicity principle in psychology has its origins in theories of perception (e.g., Hochberg & McAlister, 1953; Mach, 1959/1906). Informally, the simplicity principle states that simple explanations should be preferred – here ‘explanation’ refers to a pattern or structure in the data. The intuition is that the degree to which a pattern is suggested by the data can be quantified by assessing how briefly the data can be encoded, using that pattern.